AI Agents Become Enterprises’ Fastest Growing Attack Surface

The explosive growth of enterprise AI adoption has created a new and rapidly expanding vulnerability: AI identities and agents are now the fastest-growing attack surface for businesses, according to a warning issued in the Sophos AI Security 2026 Report, released July 22.

As employees increasingly deploy coding agents, agentic AI assistants, and large language models to boost productivity, these tools are routinely granted privileged access to core enterprise systems. This convenience, however, has opened a dangerous door. Cybercriminals are now targeting the trust, credentials, and permissions surrounding these AI systems, turning them into a high-value target for malicious actors.

A breach of an AI identity can provide attackers with a direct pathway into an enterprise network. This makes OAuth tokens, AI service credentials, developer tools, and exposed AI infrastructure prime targets. Yet, governance and security policies have not kept pace. Sophos warned that the “identity fabric connecting AI services to enterprise systems creates exposure that existing governance was not designed to handle.”

The scale of the problem is staggering. Research from BeyondTrust, cited in the Sophos report, reveals a 466.7% increase in active AI agents within enterprise environments over the past year alone.

This rapid adoption, often without proper oversight, creates an AI governance gap. Organizations risk exposure when AI tools and enterprise platforms are secured with weak passwords. Cybercriminals target workplace identities for a range of purposes, from data theft to deploying ransomware, as Sophos has previously highlighted. The elevated privileges assigned to many AI agents mean that compromising them could grant attackers additional access for broader attacks.

There is also the subtle but dangerous risk of AI tool manipulation or poisoning. Attackers could subtly direct these systems to perform actions that benefit them while harming the victim organization. Meanwhile, threat actors are actively using AI to enhance their own campaigns, including phishing, social engineering, and malware development.

“This report makes clear that AI security is no longer just about model behavior or speculative future risks,” said John Peterson, CTO at Sophos. “AI is actively being absorbed into criminal workflows and social engineering operations, as well as into enterprise software development and identity systems within legitimate organizations. That means the threat is in the here and now.”

He added that the next few months will be defined by how quickly organizations can govern AI use, secure the identities and connections around it, and keep pace with attackers who are rapidly adopting new capabilities.

To counter these threats, the report recommends treating AI agents like human users. Their access should be restricted to only the applications and services they absolutely need. Manual verification should be required for any new access to a different area, application, or service. Finally, organizations should set up alerts to detect suspicious behavior or unexpected data exfiltration by AI agent identities.